Fast CT Image Processing using Parallelized Non-local Means

Reducing the radiation dose delivered to patients has been an important concern since the introduction of X-ray computed tomography (CT). However, low-dose CT images tend to be severely degraded by noise. This paper proposes using parallelized non-local means (PNM) under a computation framework for improving low-dose X-ray CT images. For the proposed PNM method, the pixel intensities are processed based on the self-similarity properties of tissues with various levels of attenuation across large-scale neighborhoods. In the experiment, CT images from a Siemens CT scanner with 16 detector rows are collected for various dose levels. Results on both phantom and clinical CT images from various human parts validate the performance of the proposed accelerated parallel approach in terms of noise and artifact suppression and feature preservation.